chore: merge main into codex/audit-esp-zero-byte-partitions

src/artifacts/EspFlashImageArtifactReader.test.ts

@@ -175,6 +175,27 @@ describe("ESP flash layout inventory", () => {
   });
 });
 
+it("inventories a zero-byte partition without reporting missing bytes", async () => {
+  const dump = espFlashDump({
+    bootloader: { at: 0, bytes: BOOTLOADER },
+    table: partitionTable([
+      ...TWO_OTA_PARTITIONS,
+      { label: "empty", type: 1, subtype: 0x82, offset: 0xa000, size: 0 },
+    ]),
+  });
+  const snapshot = await scanCanonicalArtifactInventory(await writeDump(dump));
+  expect(snapshot.coverage.status).toBe("complete");
+  expect(
+    snapshot.occurrences.find(
+      ({ logical_path }) => logical_path === "partitions/empty.bin",
+    ),
+  ).toMatchObject({
+    declared_size: 0,
+    limitations: ["partitions/empty.bin declares zero bytes."],
+    format_metadata: { flash_address: 0xa000, size: 0 },
+  });
+});
+
 describe("ESP flash layout coverage and selection", () => {
   it("keeps usable partitions beside a truncated and a missing one", async () => {
     const cut = DUMP.subarray(0, 0x2_8000);

src/artifacts/EspFlashImageArtifactReader.ts

@@ -272,15 +272,17 @@ const regionEntry = (
   declaredIntegrity: null,
   unpacked: false,
   limitations:
-    region.available === 0
-      ? [
-          `${region.path} lies outside the dump: the file holds none of its ${String(region.length)} bytes.`,
-        ]
-      : region.available < region.length
+    region.availability === "empty"
+      ? [`${region.path} declares zero bytes.`]
+      : region.availability === "missing"
         ? [
-            `${region.path} is truncated: the dump holds ${String(region.available)} of its ${String(region.length)} bytes.`,
+            `${region.path} lies outside the dump: the file holds none of its ${String(region.length)} bytes.`,
           ]
-        : [],
+        : region.availability === "truncated"
+          ? [
+              `${region.path} is truncated: the dump holds ${String(region.available)} of its ${String(region.length)} bytes.`,
+            ]
+          : [],
   adapterKey: `esp-flash-image:${region.path}`,
   ...(region.partition === undefined
     ? {}

src/artifacts/pyinstaller/MarshalSubset.test.ts

@@ -20,11 +20,6 @@ it("reads a Python 3.9 reference-flagged PYZ table", () => {
   ]);
 });
 
-it("reads marshal one-byte string tags as Latin-1", () => {
-  for (const hex of ["6101000000e9", "4101000000e9", "7a01e9", "5a01e9"])
-    expect(readMarshalSubset(bytes(hex))).toEqual({ ok: true, value: "é" });
-});
-
 it("decodes CPython marshal Unicode with surrogatepass", () => {
   expect(
     readMarshalSubset(

src/artifacts/pyinstaller/MarshalSubset.ts

@@ -168,8 +168,14 @@ const readAsciiBytes = (
 ): Result<string, PyInstallerFailure> => {
   const bytes = readBytes(cursor, length);
   if (!bytes.ok) return bytes;
-  // These tags use CPython's one-byte Unicode constructor, not an ASCII check.
-  return ok(bytes.value.toString("latin1"));
+  if (bytes.value.some((byte) => byte > 0x7f))
+    return err(
+      pyInstallerFailure(
+        "format",
+        "PYZ marshal ASCII string contains a non-ASCII byte",
+      ),
+    );
+  return ok(bytes.value.toString("ascii"));
 };
 
 const readUtf8 = (

src/domain/firmware/espFlashLayout.ts

@@ -38,6 +38,8 @@ export interface EspFlashRegion {
   readonly fileOffset: number;
   /** Bytes of the region the file holds, from `fileOffset`. */
   readonly available: number;
+  /** An empty declaration has no bytes to be missing from the dump. */
+  readonly availability: "empty" | "missing" | "truncated" | "complete";
   readonly partition: EspPartition | undefined;
 }
 
@@ -117,12 +119,21 @@ const partitionPaths = (
 
 const region = (
   input: EspFlashPlanInput,
-  fields: Omit<EspFlashRegion, "fileOffset" | "available">,
+  fields: Omit<EspFlashRegion, "fileOffset" | "available" | "availability">,
 ): EspFlashRegion => {
   const fileOffset = fields.flashAddress - input.origin.value;
   const start = Math.max(0, fileOffset);
   const end = Math.min(input.fileSize, fileOffset + fields.length);
-  return { ...fields, fileOffset, available: Math.max(0, end - start) };
+  const available = Math.max(0, end - start);
+  const availability =
+    fields.length === 0
+      ? "empty"
+      : available === 0
+        ? "missing"
+        : available < fields.length
+          ? "truncated"
+          : "complete";
+  return { ...fields, fileOffset, available, availability };
 };
 
 const rangeText = (address: number, length: number): string =>
@@ -211,11 +222,15 @@ export const planEspFlashLayout = (input: EspFlashPlanInput): EspFlashPlan => {
       limitations.push(
         `Partition ${name} at ${offsetText(part.offset)} begins before the dump's first byte at flash ${offsetText(input.origin.value)}; it is not read.`,
       );
-    else if (placed.available === 0)
+    else if (placed.availability === "empty")
+      limitations.push(
+        `Partition ${name} at ${offsetText(part.offset)} declares zero bytes.`,
+      );
+    else if (placed.availability === "missing")
       limitations.push(
         `Partition ${name} at ${rangeText(part.offset, part.size)} lies outside the dump, which ends at flash ${offsetText(input.origin.value + input.fileSize)}.`,
       );
-    else if (placed.available < part.size)
+    else if (placed.availability === "truncated")
       limitations.push(
         `Partition ${name} at ${rangeText(part.offset, part.size)} is truncated: the dump holds ${String(placed.available)} of its ${String(part.size)} bytes.`,
       );
@@ -268,6 +283,7 @@ const overlaps = (regions: readonly EspFlashRegion[]): string[] => {
   // The region reaching furthest so far, so a long one shows every overlap.
   let furthest: EspFlashRegion | undefined;
   for (const current of ordered) {
+    if (current.availability === "empty") continue;
     if (
       furthest !== undefined &&
       current.flashAddress < regionEnd(furthest) &&

src/domain/firmware/espFlashPlan.test.ts

@@ -179,6 +179,38 @@ describe("ESP flash layout plans", () => {
 });
 
 describe("ESP flash layout overlaps", () => {
+  it.each([
+    ["inside another partition", 0xa000],
+    ["at the end of the dump", 0x4_0000],
+  ])(
+    "retains a zero-byte partition %s as an empty declaration",
+    (_, offset) => {
+      const plan = planEspFlashLayout(
+        input({
+          table: table([
+            {
+              label: "data",
+              type: 1,
+              subtype: 2,
+              offset: 0x9000,
+              size: 0x4000,
+            },
+            { label: "empty", type: 1, subtype: 0x82, offset, size: 0 },
+          ]),
+        }),
+      );
+      expect(plan.regions.at(-1)).toMatchObject({
+        path: "partitions/empty.bin",
+        fileOffset: offset,
+        length: 0,
+        available: 0,
+      });
+      expect(plan.limitations).toEqual([
+        `Partition empty at 0x${offset.toString(16)} declares zero bytes.`,
+      ]);
+    },
+  );
+
   it("reports every region inside a longer one", () => {
     const plan = planEspFlashLayout(
       input({

src/python/PythonPycParser.test.ts

@@ -407,30 +407,6 @@ it("charges long digits to the string budget and keeps large previews bounded",
   );
 });
 
-it("reads marshal one-byte string tags as Latin-1 with Unicode identity", () => {
-  for (const tag of [0x61, 0x41, 0x7a, 0x5a]) {
-    const encoded = Buffer.concat([
-      Buffer.from([tag]),
-      tag === 0x61 || tag === 0x41 ? u32(1) : Buffer.from([1]),
-      Buffer.from([0xe9]),
-    ]);
-    const sameUnicode = marshalUnicode(Buffer.from("é", "utf8"));
-    const set = Buffer.concat([
-      Buffer.from([0x3c]),
-      u32(2),
-      encoded,
-      sameUnicode,
-    ]);
-    const report = root(parse(withConsts([encoded, set])));
-    expect(report.consts[0]).toMatchObject({
-      kind: "str",
-      value: "é",
-      bytes: 1,
-    });
-    expect(report.consts[1]).toEqual({ kind: "set", length: 1 });
-  }
-});
-
 it("decodes marshal strings with CPython surrogatepass rules", () => {
   const report = parse(
     withConsts([marshalUnicode(Buffer.from("efbbbf73eda080efbbbf", "hex"))]),
@@ -1558,36 +1534,6 @@ it("hashes a long TYPE_REF tuple chain without recursing", () => {
   expect(pythonPycSchema.safeParse(report).success).toBe(true);
 });
 
-it("hashes shallow TYPE_REF code chains without using the JavaScript stack", () => {
-  const links = 3000;
-  const codes = Array.from({ length: links }, (_, index) =>
-    marshalCode("py311", {
-      flagged: true,
-      name: "f",
-      consts: index === 0 ? [Buffer.from([0x4e])] : [marshalRef(index - 1)],
-    }),
-  );
-  const bytes = Buffer.concat([
-    pythonPycHeader(3531),
-    marshalCode("py311", {
-      constsObject: Buffer.concat([
-        Buffer.from([0x28]),
-        u32(links + 1),
-        ...codes,
-        Buffer.from([0x3c]),
-        u32(1),
-        marshalRef(links - 1),
-      ]),
-    }),
-  ]);
-  const report = parseWith(bytes, { max_objects: 65_536 });
-  expect(root(report).consts.at(-1)).toEqual({ kind: "set", length: 1 });
-  expect(report.coverage).toMatchObject({
-    status: "complete",
-    code_objects: links + 1,
-  });
-});
-
 it("rejects code CPython treats as corrupt instead of complete coverage", () => {
   const file = (
     layout: "py37" | "py38" | "py311",

src/python/PythonPycParser.ts

@@ -332,7 +332,7 @@ class MarshalReader {
   readonly #codeKeys = new WeakMap<CodeNode, string>();
   readonly #codeIdentities = new WeakMap<CodeNode, number>();
   readonly #codeEquality = new WeakMap<CodeNode, CodeEquality>();
-  /** Completed Python hashability checks per code object. */
+  /** Python hashability per code object. False while that check is in progress. */
   readonly #codeHashable = new WeakMap<CodeNode, boolean>();
   /** Tuples whose items are still being read; CPython cannot hash them yet. */
   readonly #readingTuples = new WeakSet<MarshalValue>();
@@ -436,19 +436,13 @@ class MarshalReader {
         };
       case TYPE_INTERNED:
       case TYPE_UNICODE:
-        return this.string(
-          this.nonNegativeInt32("Marshal string length"),
-          "utf8",
-        );
+        return this.utf8(this.nonNegativeInt32("Marshal string length"), false);
       case TYPE_ASCII:
       case TYPE_ASCII_INTERNED:
-        return this.string(
-          this.nonNegativeInt32("Marshal string length"),
-          "latin1",
-        );
+        return this.utf8(this.nonNegativeInt32("Marshal string length"), true);
       case TYPE_SHORT_ASCII:
       case TYPE_SHORT_ASCII_INTERNED:
-        return this.string(this.u8(), "latin1");
+        return this.utf8(this.u8(), true);
       case TYPE_TUPLE:
         return this.sequenceValue(
           "tuple",
@@ -1315,70 +1309,13 @@ class MarshalReader {
   #isCodeHashable(node: CodeNode): boolean {
     const known = this.#codeHashable.get(node);
     if (known !== undefined) return known;
-    const frames: { value: MarshalValue; finish: boolean }[] = [
-      { value: { kind: "code", value: node }, finish: false },
-    ];
-    const gray = new Set<MarshalValue | CodeNode>();
-    const complete = new Set<MarshalValue>();
-    const reject = (): false => {
-      this.#codeHashable.set(node, false);
-      return false;
-    };
-    while (frames.length > 0) {
-      this.#takeKeyStep();
-      const frame = frames.pop()!;
-      const value = frame.value;
-      const identity = value.kind === "code" ? value.value : value;
-      if (frame.finish) {
-        gray.delete(identity);
-        if (value.kind === "code") {
-          const material =
-            this.#codeEquality.get(value.value) ??
-            malformed("Code object is missing equality material.");
-          // Child code checks are now complete, so the existing tuple key
-          // walker preserves its equality and expansion budgets without recursion.
-          const hashable =
-            this.#equalityKey(material.constsTuple, "hash") !== null;
-          this.#codeHashable.set(value.value, hashable);
-          if (!hashable) return reject();
-        } else complete.add(value);
-        continue;
-      }
-      if (value.kind === "code") {
-        const cached = this.#codeHashable.get(value.value);
-        if (cached === false) return reject();
-        if (cached === true) continue;
-      } else {
-        const cached = this.#keyCache.get(value);
-        if (cached === null) return reject();
-        if (cached !== undefined || complete.has(value)) continue;
-      }
-      if (
-        gray.has(identity) ||
-        this.#readingTuples.has(value) ||
-        value.kind === "list" ||
-        value.kind === "set" ||
-        value.kind === "dict" ||
-        value.kind === "null"
-      )
-        return reject();
-      const children =
-        value.kind === "code"
-          ? [
-              (
-                this.#codeEquality.get(value.value) ??
-                malformed("Code object is missing equality material.")
-              ).constsTuple,
-            ]
-          : value.kind === "tuple" || value.kind === "frozenset"
-            ? value.items
-            : [];
-      gray.add(identity);
-      frames.push({ value, finish: true });
-      for (let index = children.length - 1; index >= 0; index -= 1)
-        frames.push({ value: children[index]!, finish: false });
-    }
-    return true;
+    this.#codeHashable.set(node, false);
+    const material =
+      this.#codeEquality.get(node) ??
+      malformed("Code object is missing equality material.");
+    const hashable = this.#equalityKey(material.constsTuple, "hash") !== null;
+    this.#codeHashable.set(node, hashable);
+    return hashable;
   }
 
   #codeIdentity(node: CodeNode): number {
@@ -1555,20 +1492,17 @@ class MarshalReader {
     );
   }
 
-  string(length: number, encoding: "latin1" | "utf8"): MarshalValue {
+  utf8(length: number, ascii: boolean): MarshalValue {
     const bytes = this.blob(length);
-    if (encoding === "latin1") {
-      // CPython reads the ASCII-tagged forms as one-byte Unicode, even when
-      // high bytes appear. Identity must agree with the UTF-8 string form.
-      const value = bytes.toString("latin1");
+    if (ascii) {
+      if (bytes.some((byte) => byte > 0x7f))
+        malformed("Marshal ASCII string contains a non-ASCII byte.");
       return {
         kind: "str",
-        value,
+        value: bytes.toString("latin1"),
         previewEnd: Math.min(length, PYTHON_PYC_CONST_PREVIEW_LIMIT),
         bytes: length,
-        raw: bytes.some((byte) => byte > 0x7f)
-          ? Buffer.from(value, "utf8")
-          : bytes,
+        raw: bytes,
       };
     }
     const decoded = decodeMarshalUtf8(bytes, PYTHON_PYC_CONST_PREVIEW_LIMIT);